parallel restore

Attached is the latest version.

Changes:

. some tidy up as variously requested.
. some common code is factored out
. some descriptive comments added
. platform specific stuff (e.g. spawn, reap) is factored out
. --truncate_before_load is gone, and we now do this during parallel
restore if we created the table. For a non-parallel restore the
equivalent would be to run the whole restore in a single transaction.

One of the hardest parts of getting this to work was handling
dependencies right. I will work on adding some more comments regarding that.

Simon asked about a way to adjust the number of worker children as we go
along. That's way out of scope at this stage. In testing it appears that
the sweet spot is roughly m=number_of_processors, which makes some
sense, but more experience will clarify this.

cheers

andrew